Parallel Tcad Optimization and Parameter Extraction for Computationally Expensive Objective Functions

The SIESTA (Simulation Environment for Semiconductor Technology Analysis) framework is an extensible tool for optimization and inverse modeling of semiconductor devices including dynamic load balancing for taking advantage of several, loosely connected workstations. Because of the increasing computational power available today, the use of evolutionary computation optimizers which usually require a large number of evaluations of the objective functions becomes feasible even for problems with computationally very expensive objective functions. After a brief introduction to the SIESTA framework and its capabilities, we compare the performance of its optimizers at a real world parameter extraction problem and find that for certain problems genetic algorithms and simulated annealing perform better than gradient based optimization.